Neural machine translation (DeepL, Google Translate, the engines inside Trados, memoQ, and Phrase) has been in your workflow for years. What changed is that large language models translate with context: give Claude or ChatGPT a glossary, a style guide, and three approved translations, and the draft comes back in the client's register with terminology mostly right. That has moved a lot of paid work from translation to post-editing, and pushed rates down for anything a client believes a machine can do.
The work that holds its value is the work where being wrong costs something: contracts, patents, regulatory filings, clinical documents, marketing that has to land in a culture, literary voice, and anything that needs a certified or sworn translation. Interpreting is changing more slowly. Live speech translation in Zoom, Teams, and Google Meet is usable for casual meetings and not acceptable for depositions, medical encounters, or negotiations where a mistranslated modal verb matters.
Use AI to draft, to check consistency, to research terminology, and to build the glossaries and translation memories that make you faster. Keep the final read, the certification, the confidentiality decisions, and the conversation with clients about what they are buying.
Quick wins this week
- Build a glossary for your biggest client: paste three approved translations into a chat assistant, ask it to extract term pairs, then review and import them into your CAT tool.
- Run one low-stakes document through DeepL and through a chat assistant loaded with your glossary, and compare where each fails; that tells you which to use for what.
- Paste source and target of a finished job and ask for mistranslations, omissions, number and date mismatches, and inconsistent terms: the QA pass you skip when rushed.
- Before an interpreting assignment, feed the agenda, bios, and deck into NotebookLM and generate a bilingual term list.
What AI can do for translators, task by task
Machine translation post-editing
Run the source through a chat assistant with the glossary, style notes, and an approved sample, compare its draft with the engine's, and post-edit the better one. Check every number, date, unit, name, negation, and legal modal (shall, may, must) against the source, because fluent output hides errors exactly there. Agree with the client whether they are paying for light or full post-editing; ISO 18587 defines both.
Terminology research and glossary building
Ask a chat assistant to extract term pairs from approved translations, then verify each candidate against authoritative sources: IATE for EU terminology, the regulator's own glossary, the client's website, the standard itself. Models invent plausible renderings for niche terms and pick the wrong one of two accepted forms, so treat suggestions as leads. Save only verified pairs in your termbase.
Consistency and QA on long documents
Paste source and target into Claude, which handles long texts well, and ask for a segment-by-segment report of meaning changes, omissions, number mismatches, and inconsistent terms, quoting both sides. Still run your CAT tool's QA for tags and placeholders. The model catches what tired eyes miss; it also flags things that are fine, so judge each item.
Transcreation and marketing copy
Ask for three versions of a tagline or campaign line: close, adapted, and rewritten for intent, with a back-translation and notes on cultural or legal references to check. Models default to a neutral 'international' register, so specify the market, the audience, and the brand voice with examples. You choose; the client sees the back-translation.
Interpreting preparation
Load the agenda, speaker bios, decks, and prior minutes into NotebookLM or Gemini and ask for a bilingual term list, the likely issues, and each side's positions. Practice names and numbers with text-to-speech. Live AI captions can back you up in a booth for a term you missed; they cannot replace you where a misheard 'not' changes an outcome.
Subtitling, dubbing, and audio localization
ElevenLabs and HeyGen dub video into other languages in the speaker's own voice, and a chat assistant translates subtitle files within character-per-second limits. Review every line for timing, register, and mistranslated idioms, and confirm the client has the speaker's consent for a cloned voice before you deliver anything.
Scanned and handwritten source documents
For certified work on birth certificates, diplomas, and court records, Gemini or ChatGPT can read a scan or photo and produce a text version to translate from. OCR errors cluster in names, dates, seals, and stamps, so check every field against the image and describe illegible portions as illegible rather than guessing.
Prompts for translators
Replace the bracketed placeholders, paste into any chat assistant, and iterate on the result.
Context-loaded translation draft
You are a professional [SOURCE LANGUAGE] to [TARGET LANGUAGE] translator specializing in [DOMAIN, E.G. COMMERCIAL CONTRACTS]. Translate the text below for [CLIENT AND PURPOSE]. Follow the glossary exactly, match the register of the sample, and preserve all numbers, dates, names, and formatting tags. Do not summarize, omit, or add. After the translation, list every passage where you were uncertain, the alternatives you considered, and any source-text ambiguity I should raise with the client. Glossary: [PASTE TERM PAIRS] Sample: [PASTE AN APPROVED TRANSLATION] Source: [PASTE SOURCE TEXT]
Tip: The uncertainty list is where your post-editing time should go first.
Bilingual QA pass
Act as a senior reviser. Compare the source and target below segment by segment. Report in a table with segment numbers: mistranslations, omissions, additions, terminology that conflicts with the glossary, number, date, unit, or currency mismatches, wrong negations or modals, untranslated text, and tag errors. Do not rewrite the translation; quote both segments and rank issues by severity. If a category has no findings, say so. Glossary: [PASTE GLOSSARY] Source: [PASTE SOURCE] Target: [PASTE TARGET]
Tip: Run it before your own final read, not instead of it.
Term research with sources to verify
You are a terminologist. For the [SOURCE LANGUAGE] term '[TERM]' as used in [CONTEXT, E.G. A GERMAN EMPLOYMENT CONTRACT], give the candidate [TARGET LANGUAGE] equivalents used in [JURISDICTION OR INDUSTRY], the register and legal or technical connotation of each, which one official or regulatory sources use, and the false friends to avoid. Tell me what kind of authoritative source I should verify each in (statute, standard, regulator glossary). If you are not confident a rendering is real usage, say so rather than guess.
Tip: Verify before the term enters your termbase; a wrong entry repeats forever.
Interpreting assignment prep
You are preparing me to interpret [MODE, E.G. CONSECUTIVE] between [LANGUAGE A] and [LANGUAGE B] at [ASSIGNMENT, E.G. A SUPPLIER NEGOTIATION ABOUT PACKAGING MACHINERY]. From the materials below build a bilingual term list of technical vocabulary, acronyms, and product names; the participants with titles and pronunciation notes; a one-page summary of the issues and each side's likely positions; and ten phrases in each language for procedural moments (asking a speaker to pause, requesting clarification). Flag terms with more than one accepted rendering. Materials: [PASTE AGENDA, BIOS, AND DECK TEXT]
Tip: Ask for the term list as a two-column table you can print for the booth.
Explain the service tiers to a client
Write a reply to the client email below in [LANGUAGE], warm and professional, under 200 words. Explain the difference between raw machine translation, post-edited machine translation, full human translation, and a certified translation for [PURPOSE, E.G. A USCIS SUBMISSION], and which I recommend for their document and why. State what my certified translation includes (a signed accuracy statement) and does not (notarization, legal advice). Ask the three questions I need to quote: word count and format, deadline, and intended use. Client email: [PASTE EMAIL]
Tip: Save the answer as a template; you will send some version of it weekly.
Want a prompt for something else? Use the Prompt Builder.
Skills to build
Post-editing at speed without lowering the bar
Why: Post-editing is where the volume is, and it pays only if you are faster than translating from scratch while catching the errors fluent output hides.
How: Time yourself on ten jobs, log the error types you find, and build a personal checklist of what to check first (numbers, negations, modals, terms).
Glossary and translation-memory discipline
Why: Context is what makes a model's draft usable, and your glossaries and memories are the context nobody else has.
How: Keep a termbase per client, extract pairs from every approved job, and paste the relevant glossary into every prompt.
Confidentiality triage
Why: Which tool a document can go into depends on the client's NDA, the data in it, and the tool's retention terms, and the wrong call ends a relationship.
How: Sort clients into tiers (public material, NDA, regulated) and write down which tools each tier allows. Get written approval where you are unsure.
Explaining what clients are buying
Why: Clients cannot tell machine output from your work until it fails, so your job includes explaining the tiers and the risks in their language.
How: Write a one-page service menu with plain definitions of machine translation, post-editing, human translation, and certified translation, and what each is appropriate for.
Specializing
Why: Generalist translation is the work machines took; legal, medical, technical, and creative specialization is what clients still pay a person for.
How: Pick one domain, read its source texts the way its professionals do, and build the termbase that makes you the fastest and safest choice in it.
Tools worth knowing
Claude
A careful writing and analysis assistant that shines on long documents.
ChatGPT
The general-purpose AI assistant most of your coworkers already use.
Gemini
Google's assistant, strongest when your work already lives in Google Workspace.
NotebookLM
A research notebook that only answers from the sources you give it, with citations.
ElevenLabs
Lifelike AI voices for narration, dubbing, transcription, and voice agents
HeyGen
AI avatar videos and lip-synced translation for training, sales, and marketing.
Cautions for translators
Free translation sites and consumer chat tools may retain what you paste and use it for training, which can breach an NDA by itself. Medical records fall under HIPAA, legal documents under attorney-client privilege, EU personal data under GDPR. Never paste confidential documents into a consumer AI tool unless the client or your agency has approved it; use professional plans with no-training terms and deletion controls, anonymize where you can, and work inside the client's environment when they require it.
In the US a certified translation is one you sign an accuracy statement for, and USCIS requires one for foreign-language documents; some countries require sworn translators appointed by courts. A machine cannot certify anything, and you should never certify output you have not reviewed word by word, because your signature carries the liability. ATA certification is a credential you hold; certifying a translation is a statement you make about one document.
Model errors are quiet: a dropped 'not', a shifted number, a 'may' that became 'must', a defined term that drifts halfway through a long document, a gendered noun assigned by stereotype. Check the categories that carry risk in every job rather than reading for flow.
Real-time speech translation is fine for a casual call and unfit for courts, medical encounters, immigration interviews, and negotiations; it cannot ask for clarification, hears accents unevenly, and misses register. Language-access rules in healthcare and courts still assume a qualified human. Use it as a backup for a missed term, never as the interpreter.
Some clients prohibit AI use and others expect it; find out before the job and keep records of how each was done. Do not bill full human rates for lightly reviewed machine output, and do not accept post-editing rates for work you had to retranslate.
Your 30-day plan
- Week 1: Read the data terms of every tool you use and sort your clients into confidentiality tiers. Ask for written approval where an NDA is unclear.
- Week 2: Build glossaries for your three biggest clients from approved translations and save a context-loaded translation prompt for each.
- Week 3: On real jobs, compare the engine's draft with a chat assistant's, time your post-editing, and log the error types you catch.
- Week 3: Run the bilingual QA prompt on every delivery before your final read.
- Week 4: Write your service menu and certified-translation statement template, update your rate card, and, if you interpret, run the prep workflow on your next assignment.
Frequently asked questions
Will AI replace translators and interpreters?
Is it safe to use ChatGPT or DeepL for confidential documents?
Can AI produce a certified translation?
Which is better for translation, DeepL or ChatGPT?
Terms used on this page
Related roles
- AI for Content WritersAI can research, outline, draft, and edit alongside you. The writers who thrive with it use it to reach the interesting part faster, keep their own voice, and are straight with clients about how they work.
- AI for Technical WritersAssistants draft reference docs from code, turn engineer notes into procedures, and keep a docs-as-code repo tidy through a coding agent. They also invent flags and parameters that do not exist, so verification against the running product is the job.
- AI for ParalegalsParalegals live in documents, deadlines, and details. AI can summarize, organize, and draft at speed, as long as client material stays inside firm-approved tools and every citation gets checked in Westlaw or Lexis before anyone relies on it.
- AI for Customer Service RepsLet AI draft the reply, find the policy, and summarize the thread so you can spend your energy on the customer in front of you. Here is what to hand off, what to check, and what never goes into a chat tool.